@ruvector/ruvllm 3.0.0 → 4.0.0

package/dist/cjs/engine.d.ts

@@ -1,160 +0,0 @@
-/**
- * RuvLLM Engine - Main orchestrator for self-learning LLM
- */
-import { RuvLLMConfig, GenerationConfig, QueryResponse, RoutingDecision, MemoryResult, MemoryId, RuvLLMStats, Feedback, Embedding, BatchQueryRequest, BatchQueryResponse, ChatMessage, GenerationResult, LoadModelOptions, LoadedModelInfo } from './types';
-/** Returned by generate()/query() only with `allowPlaceholder` / `RUVLLM_ALLOW_MOCK=1`. */
-export declare const PLACEHOLDER_TEXT = "[RuvLLM placeholder: no language model is loaded; this text is not model output]";
-/**
- * RuvLLM - Self-learning LLM orchestrator
- *
- * Combines SONA adaptive learning with HNSW memory, FastGRNN routing, and
- * GGUF inference through the native candle backend.
- *
- * Text comes only from a model loaded with `loadModel()` (or the `modelPath`
- * option). Without one, `generate()`, `query()`, `chat()` and
- * `generateDetailed()` throw `RUVLLM_NO_LANGUAGE_MODEL`.
- *
- * @example
- * ```typescript
- * import { RuvLLM } from '@ruvector/ruvllm';
- *
- * const llm = new RuvLLM();
- * llm.loadModel('./qwen2.5-0.5b-instruct-q4_k_m.gguf');
- *
- * const out = llm.chat([{ role: 'user', content: 'What is machine learning?' }], {
- *   maxTokens: 128,
- * });
- * console.log(out.text, out.finishReason, out.completionTokens);
- *
- * // Routed query, answered by the loaded model
- * const response = llm.query('What is machine learning?');
- * llm.feedback({ requestId: response.requestId, rating: 5 });
- * ```
- */
-export declare class RuvLLM {
-    private native;
-    private nativeVersion;
-    private config;
-    private allowPlaceholder;
-    private fallbackState;
-    /**
-     * Create a new RuvLLM instance
-     */
-    constructor(config?: RuvLLMConfig);
-    /**
-     * Whether the native binary can load and run a language model. False
-     * without a native binary, for binaries that predate `loadModel` (the 2.0.x
-     * `@ruvector/ruvllm-<platform>` packages), and for binaries built without
-     * the `candle` inference backend.
-     */
-    supportsModelLoading(): boolean;
-    /** Error for a binary that has `loadModel` but no inference backend. */
-    private noBackendError;
-    /**
-     * Load a GGUF model (or a directory holding one) for `generate()`,
-     * `chat()` and `query()`. Supported GGUF architectures: llama, mistral,
-     * qwen2, qwen3; others fail with an error naming the architecture. The
-     * tokenizer comes from a `tokenizer.json` beside the file, else from the
-     * GGUF itself. Replaces any loaded model; if loading fails, no model is
-     * loaded. Synchronous: blocks the event loop while the weights load.
-     *
-     * @throws `RUVLLM_NATIVE_UNAVAILABLE` without a native binary,
-     *   `RUVLLM_NATIVE_TOO_OLD` when the binary predates model loading,
-     *   `RUVLLM_NO_INFERENCE_BACKEND` when it was built without `candle`,
-     *   `RUVLLM_MODEL_NOT_FOUND` / `RUVLLM_MODEL_LOAD_FAILED` from the backend.
-     */
-    loadModel(path: string, options?: LoadModelOptions): LoadedModelInfo;
-    /**
-     * Unload the language model and free its memory
-     */
-    unloadModel(): void;
-    /**
-     * Whether a language model is loaded
-     */
-    isModelLoaded(): boolean;
-    /**
-     * The loaded language model, or null
-     */
-    modelInfo(): LoadedModelInfo | null;
-    /**
-     * Route `text`, then answer it with the loaded model (as one user message
-     * under the model's chat template).
-     *
-     * @throws `RUVLLM_NO_LANGUAGE_MODEL` when no model is loaded, unless
-     *   `allowPlaceholder` is set (then `text` is `PLACEHOLDER_TEXT`).
-     */
-    query(text: string, config?: GenerationConfig): QueryResponse;
-    /**
-     * Continue `prompt` verbatim with the loaded model (no chat template; use
-     * `chat()` for instruction-tuned models). Returns the generated text.
-     *
-     * @throws `RUVLLM_NO_LANGUAGE_MODEL` when no model is loaded, unless
-     *   `allowPlaceholder` is set (then it returns `PLACEHOLDER_TEXT`).
-     */
-    generate(prompt: string, config?: GenerationConfig): string;
-    /**
-     * Like `generate()`, with token counts (from the model's tokenizer) and
-     * the finish reason. Always throws `RUVLLM_NO_LANGUAGE_MODEL` without a
-     * loaded model.
-     */
-    generateDetailed(prompt: string, config?: GenerationConfig): GenerationResult;
-    /**
-     * Run the loaded model on a conversation, formatted with the model's chat
-     * template (ChatML for Qwen, Llama 3 / Mistral formats for those models).
-     * Always throws `RUVLLM_NO_LANGUAGE_MODEL` without a loaded model.
-     */
-    chat(messages: ChatMessage[], config?: GenerationConfig): GenerationResult;
-    private requireModel;
-    private noModelError;
-    /** Throw `RUVLLM_NO_LANGUAGE_MODEL`, or return the labelled placeholder when opted in. */
-    private placeholderOrThrow;
-    /**
-     * Get routing decision for a query
-     */
-    route(text: string): RoutingDecision;
-    /**
-     * Search memory for similar content
-     */
-    searchMemory(text: string, k?: number): MemoryResult[];
-    /**
-     * Add content to memory
-     */
-    addMemory(content: string, metadata?: Record<string, unknown>): MemoryId;
-    /**
-     * Provide feedback for learning
-     */
-    feedback(fb: Feedback): boolean;
-    /**
-     * Get engine statistics
-     */
-    stats(): RuvLLMStats;
-    /**
-     * Force router learning cycle
-     */
-    forceLearn(): string;
-    /**
-     * Get embedding for text
-     */
-    embed(text: string): Embedding;
-    /**
-     * Compute similarity between two texts
-     */
-    similarity(text1: string, text2: string): number;
-    /**
-     * Check if SIMD is available
-     */
-    hasSimd(): boolean;
-    /**
-     * Get SIMD capabilities
-     */
-    simdCapabilities(): string[];
-    /**
-     * Batch query multiple prompts
-     */
-    batchQuery(request: BatchQueryRequest): BatchQueryResponse;
-    /**
-     * Check if native module is loaded
-     */
-    isNativeLoaded(): boolean;
-}
-//# sourceMappingURL=engine.d.ts.map

package/dist/cjs/engine.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"engine.d.ts","sourceRoot":"","sources":["../../src/engine.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EACL,YAAY,EACZ,gBAAgB,EAChB,aAAa,EACb,eAAe,EACf,YAAY,EACZ,QAAQ,EACR,WAAW,EACX,QAAQ,EACR,SAAS,EACT,iBAAiB,EACjB,kBAAkB,EAClB,WAAW,EAEX,gBAAgB,EAChB,gBAAgB,EAChB,eAAe,EAChB,MAAM,SAAS,CAAC;AAsHjB,2FAA2F;AAC3F,eAAO,MAAM,gBAAgB,qFACuD,CAAC;AAErF;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,qBAAa,MAAM;IACjB,OAAO,CAAC,MAAM,CAA6B;IAC3C,OAAO,CAAC,aAAa,CAAuB;IAC5C,OAAO,CAAC,MAAM,CAAe;IAC7B,OAAO,CAAC,gBAAgB,CAAU;IAGlC,OAAO,CAAC,aAAa,CAInB;IAEF;;OAEG;gBACS,MAAM,CAAC,EAAE,YAAY;IA4BjC;;;;;OAKG;IACH,oBAAoB,IAAI,OAAO;IAQ/B,wEAAwE;IACxE,OAAO,CAAC,cAAc;IAUtB;;;;;;;;;;;;OAYG;IACH,SAAS,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,eAAe;IA2BpE;;OAEG;IACH,WAAW,IAAI,IAAI;IAInB;;OAEG;IACH,aAAa,IAAI,OAAO;IAIxB;;OAEG;IACH,SAAS,IAAI,eAAe,GAAG,IAAI;IAKnC;;;;;;OAMG;IACH,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,gBAAgB,GAAG,aAAa;IA8B7D;;;;;;OAMG;IACH,QAAQ,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,gBAAgB,GAAG,MAAM;IAK3D;;;;OAIG;IACH,gBAAgB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,gBAAgB,GAAG,gBAAgB;IAO7E;;;;OAIG;IACH,IAAI,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,MAAM,CAAC,EAAE,gBAAgB,GAAG,gBAAgB;IAK1E,OAAO,CAAC,YAAY;IAKpB,OAAO,CAAC,YAAY;IA0BpB,0FAA0F;IAC1F,OAAO,CAAC,kBAAkB;IAW1B;;OAEG;IACH,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,eAAe;IAsBpC;;OAEG;IACH,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,SAAK,GAAG,YAAY,EAAE;IAmClD;;OAEG;IACH,SAAS,CAAC,OAAO,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,QAAQ;IAexE;;OAEG;IACH,QAAQ,CAAC,EAAE,EAAE,QAAQ,GAAG,OAAO;IAO/B;;OAEG;IACH,KAAK,IAAI,WAAW;IA8BpB;;OAEG;IACH,UAAU,IAAI,MAAM;IAOpB;;OAEG;IACH,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS;IAmB9B;;OAEG;IACH,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM;IA6BhD;;OAEG;IACH,OAAO,IAAI,OAAO;IAOlB;;OAEG;IACH,gBAAgB,IAAI,MAAM,EAAE;IAO5B;;OAEG;IACH,UAAU,CAAC,OAAO,EAAE,iBAAiB,GAAG,kBAAkB;IAS1D;;OAEG;IACH,cAAc,IAAI,OAAO;CAG1B"}